Modelling the seismic activity of Kahramanmaraş, Türkiye with recurrent neural network (RNN) and long short-term memory (LSTM) methods

Earthquake prediction and early warning systems play a crucial role in mitigating seismic hazards and improving disaster preparedness. Kahramanmaraş, located at the intersection of the Eastern Anatolian Fault Zone and the Dead Sea Fault Zone, is one of the most seismically active regions in Turkey. Accurately predicting earthquake magnitude and depth is essential for developing effective risk assessment strategies in such high-risk areas. This study evaluates the predictive performance of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) models using seismic data recorded between 1950 and 2023. The RNN model achieved a Mean Absolute Error (MAE) of 1.95 × 10⁻3 and an R2 value of 0.9989, outperforming the LSTM model, which yielded an MAE of 3.54 × 10⁻3 and an R2 value of 0.9957. Cross-validation results indicate that the RNN model’s Test MAE values ranged from 0.90 × 10⁻3 to 2.74 × 10⁻3, whereas the LSTM model exhibited higher deviations. These results suggest that the RNN model provides more reliable predictions for earthquake magnitude and depth. The findings highlight the potential of artificial intelligence-based models in improving seismic forecasting and early warning systems. Future research should focus on adapting these models to different seismic regions and optimizing their performance for broader applications in disaster risk reduction.